Application Of Machine Learning In High Speed Railway Optical Transport Network

The OTN(Optical Transport Network)is the guarantee for the normal operation of high-speed railways.The failure of the optical transmission network will lead to the interruption of the business and the loss of information,which will not only affect the safe operation of the train,but also cause economic losses to the user.Therefore,in the event of a network failure,an efficient fault location mechanism can provide conditions for the rapid recovery of the fault,and is also a guarantee for the safe operation of the network.In addition,the current high-speed railway information business continues to increase,and the types of services tend to be diversified.In order to meet business requirements,the topology of optical transport network is becoming more and more complex.Under this trend,how to ensure the good quality of service(QoS)of the network has become one of the most basic problems,and the congestion control of optical network is the key technology to achieve the quality of service mechanism.In this paper,machine learning algorithm is applied to fault location and congestion prediction of optical networks.The main research contents are as follows:(1)A method based on GRU(Gated Recurrent Unit)neural network for fault location is proposed.It mainly solves the single link fault location of the optical transport network.The method utilizes a neural network to train historical fault data of the link,and then locates the faulty link by analyzing the state information of the link at the time of the fault.Then,based on the backbone backbone link of the railway,a dynamic business model is built to simulate the fault location.In the simulation,the parameters of GRU are optimized,and the similar conditions are compared with the similarly trained LSTM(Long Short Term Memory)neural network,which shows the relatively excellent performance of the GRU model.The simulation results show that the proposed method can quickly locate faults,reduce the time from fault finding to repairing faults,and has 95.7%positioning accuracy,while saving the cost of using monitors.(2)A prediction model based on Support Vector Machine(SVM)for optical network congestion is proposed,which is mainly for the problem of optical transmission network congestion.The model can predict and classify the network congestion degree of the next time period in the optical network according to the information specifically allocated by the service(the size of each service bandwidth allocation,the duration of each service,the source node,and the destination node).The SVM used in the experiment is a one-to-one multivariate classification model,the kernel function is RBF,and the parameters of the SVM are optimized by the grid search method.The simulation results show that the prediction model of SVM-based optical network congestion degree has 97.8%classification accuracy,which provides a good strategy for optical network management and planning,rational allocation of optical network resources and improvement of QoS of optical network services.